Self-adaptive wavelet denoising for feature extraction of mechanical fault diagnosis based on a modified sparse coding shrinkage

A new wavelet denoising method based on a modified sparse coding shrinkage is proposed to remove noise from the signals with sparse probability density distribution. The main idea is to utilize maximum likelihood estimation of super-Gaussian signals corrupted by Gaussian noise to derive a thresholding rule, and use the wavelet soft-thresholding shrinkage on the components of sparse coding to reduce noise. In addition, a noise estimation algorithm based on signal complexity is proposed to estimate noise variance. The simulation results show that the self-adaptive denoising technique based on the sparse coding shrinkage technique is effective and efficient. This method also shows excellent performance when applied to extract abnormal features for roller bearings.